Spatiotemporal Neural Dynamics of Visual Decisions

University Of Colorado At Boulder, Boulder CO

Investigators

Abstract

An important aspect of seeing is deciding what to focus on. Such decisions are governed by activity across multiple brain areas. State of the art models of the human visual system tend to treat it as a feedforward machine, but feedback connections are also very important to the efficiency of biological vision. This project will develop mathematical models of the visual system that demonstrate how visual processing, decision making, and attention are intimately linked. These models will tested on three tasks: visually tracing a curve, visual searching a tree, and detecting motion. Results will show the degree to which the human visual system does not simply filter images, but can perform complex decision making tasks. Treatment of blindness with visual prostheses requires continued research on camera-to-brain interfaces that require deep computational knowledge of the brain's visual processing abilities. This project will develop novel and important mathematical insights into the efficiency of the visual system and the underlying neural principles. Results from this work will also be leveraged to develop course material for newly developed major and professional masters in Statistics and Data Science at CU Boulder, training the next generation of data scientists. In addition, this work will augment typical machine learning approaches to modeling human vision, and potentially inspire new technologies for computer-aided vision and categorization. Mathematical theories of vision must be extended to address cognitive tasks that reflect the complexity of the natural world. This project advances this effort for three main reasons. First, it focuses on understanding how the visual cortex participates in decision-making, a ubiquitous cognitive process. Most studies of the visual system consider object filtering and image classification. This project will develop spatially-extended neuronal network models that both process spatiotemporal inputs and interpret them to make complex decisions. These models will reflect the geometry of the visual world that humans observe and interpret, and the resulting theories will help to predict strategies people use to make everyday visual decisions. Second, the project is grounded in mathematical models of the visual cortex we will validate with the data sets of experimental collaborators. These models are amenable to mathematical analysis, so the architecture of neuronal networks can be linked to the cognitive computations they perform, providing testable predictions. This skill set will be crucial for analyzing models of spatiotemporal activity in visual cortex. Trainees will become conversant in techniques for the analysis of nonlinear and stochastic systems as well as probabilistic models of decision-making applicable to many problems in science and industry. The PI will share educational graphics, software, and outreach materials on a webpage to communicate findings widely. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.